We have a large AWS RDS running, but recently we have seen poor disk performance. We have narrowed it down to our EBSByteBalance going to 0, resulting in high latency for read & writes to the disk.

According to the docs of the underlying gp2 storage, we should have a baseline IOPS of 12k. (we have 4,000+ GB storage size)

If I check the metrics, it seems like we never hit that 12k IOPS limit. Still, our EBSByteBalance gets drained:

Am I misinterpreting the available IOPS (12k) that we have?